【摘要】 K-prototypes算法只适合处理对称标称型数据、序数型数据和区间型数据,对于用户兴趣等非对称型数据之间的相异度计算,如果采用对称标称型数据间的计算方法,则误差较大,而且设定的分类属性权重调整系数不容易确定。在考虑多种属性数据特征的基础上,对K-prototypes算法加以改进,提出一种简单的各类属性权重系数计算方法,即按属性比例初步计算各类属性权重,并分别配以微调系数进一步微调。同时扩展其算法,使其可以更好地处理非对称标称型数据,提升聚类效果。最后在实际的客户细分应用中验证其有效性。更多还原

【Abstract】 K-prototypes algorithm is only suitable for dealing with symmetric nominal data,ordinal data and interval data,for user interest asymmetric data between the dissimilarity calculations.If the calculation method of symmetrical data was used,the error was larger and setting the attribute weight adjustment coefficient was not easy to determine.On the basis of considering various attribute data characteristics,the K-prototypes algorithm was improved,and a kind of simple calculated method was put forward.All kinds of attribute weight coefficient calculation method was according to the proportion a preliminary calculation of various kinds of attribute weights,and with further fine-tuning coefficient respectively.To expand its algorithm at the same time,we can make it better to deal with asymmetric nominal data,improve the clustering effect.Finally,in the actual application of customer segmentation verify its effectiveness.更多还原